textsnap 命令

常用示例

说明

textsnap is a command-line OCR utility built around the PaddleOCR-VL-1.5 vision-language model exported to ONNX. It reads an image from a file path, a URL, or the system clipboard and writes the recognized text to a file inside ./textsnaps/, printing only the output path on stdout so it composes cleanly in shell pipelines. The model runs entirely on the CPU, no GPU and no cloud calls. Default output is Markdown so structure such as tables, headings, and lists is preserved; --plaintext flattens it for callers that only want raw text. Webpage URLs are rendered before OCR, which makes the tool usable as a "screenshot to text" pipeline for content that is hard to copy directly.

参数

-o, --output _PATH_

Write the OCR text to _PATH_. Default: ./textsnaps/_name_\_ocr.txt.

-v, --verbose

Print progress diagnostics to stderr.

--plaintext

Convert the default Markdown output into plain text (no tables, no headings).

--model-dir _PATH_

Use ONNX model files from _PATH_ instead of the cached download.

--max-tokens _N_

Cap the decoder at _N_ generated tokens (default 2048).

--max-pixels _N_

Limit the vision encoder pixel budget per image to _N_.

--no-verify

Skip SHA-256 verification of downloaded model files.

--generate-checksums

Re-download the model and rewrite the checksum manifest.